In recent years, a technique has been known by which a virtual machine (VM) that is currently working can be moved from a server in which the VM is constructed to another server without a halt. This technique is called “live migration”. During a live migration, a management server copies a memory image of the working VM onto the server at the moving destination (hereinafter, “moving destination server”). Because the contents of the memory keep changing also during the copying process, the management server copies the difference caused by the change again. After repeating this process a plurality of times, the management server finally halts the VM at the origin of the move temporarily and copies the final difference. After that, the management server copies the state of the Central Processing Unit (CPU) and the like onto the moving destination and resumes the VMs at the moving destination after the copying process.
The live migration technique is used when an administrator applies a patch to virtual software (called “Virtual Machine Monitors [VMMs]”) provided in a server. A specific example of a related VMM patch application will be explained, with reference to FIG. 18. FIG. 18 is a drawing for explaining the related VMM patch application. As illustrated in FIG. 18, in a server 1, VM 1, VM 2, and VM 3 are constructed with a VMM to which a patch has not yet been applied. To apply a patch to the VMM in the server 1, a management server moves VM 1, VM 2, and VM 3 from the server 1 to a different server (“server 2” in the present example) by executing a live migration. After that, when the VMs in the server 1 have been moved, the management server applies the patch to the VMM provided in the server 1.
Further, a method for determining a moving destination server for VMs is disclosed as another related technique. According to this method, to reduce downtime of the VMs caused by the moves, a management server determines an appropriate moving destination server by using an evaluation value that is calculated while taking into account a CPU usage percentage, a memory usage amount, and the time it takes to move the VMs, based on information collected from servers.
Patent Document 1: Japanese Laid-open Patent Publication No. 2008-217302
The related technique, however, has a problem where, when the administrator applies the patch to the VMM, it takes time for the management server to migrate the VMs grouped with the VMM to the moving destination by executing the live migration. In other words, if the management server executes the live migration while the frequency of use of the memory by the VMs is high, for example, because the contents of the memory change over a long period of time, it takes a long time to copy the difference at the final step.
Further, even if the management server determines the appropriate moving destination server, if the live migration is executed while the frequency of use of the memory by the VMs is high, the situation where the migration takes a long time remains unchanged.